Prompt tuning using diff format outputs

By using diff format space and a multi-objective multi-arm bandit algorithm, prompt tuning for large language models becomes more efficient, reducing computational resources and time, and optimizing prompts for multiple objectives effectively.

US20260141245A1Pending Publication Date: 2026-05-21MICROSOFT TECHNOLOGY LICENSING LLC
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
US · United States
Patent Type
Applications(United States)
Current Assignee / Owner
MICROSOFT TECHNOLOGY LICENSING LLC
Filing Date
2024-11-19
Publication Date
2026-05-21

AI Technical Summary

Technical Problem

Existing prompt tuning methods for large language models are computationally intensive and time-consuming, particularly when dealing with large prompts, and often require significant human effort and expertise, leading to inefficiencies in optimizing prompts for multiple objective targets.

Method used

Implementing prompt tuning techniques that operate in diff format space to generate and optimize prompt candidates, utilizing a multi-objective multi-arm bandit algorithm to select the best performing prompt variant based on multiple target metrics, and applying natural language gradients to iteratively refine prompts.

Benefits of technology

This approach significantly reduces computational resources and inference time, enabling faster prompt tuning and optimization across multiple objective targets, resulting in more efficient and automated prompt development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US20260141245A1-D00000_ABST
    Figure US20260141245A1-D00000_ABST
Patent Text Reader

Abstract

Implementations using diff formats for multi-objective prompt tuning are provided. One implementation includes a computing system comprising processing circuitry and memory storing instructions that, during execution, causes the processing circuitry to receive an initial prompt, generate a plurality of prompt variants based on the initial prompt, wherein each of the prompt variants is in a diff format that describes changes from the initial prompt, derive a plurality of prompt candidates based on the initial prompt and the plurality of prompt variants, wherein each of the prompt candidates is derived by applying the changes described in a respective prompt variant to the initial prompt, evaluate the plurality of prompt candidates to determine a quality of each prompt candidate based on at least one target metric, and select and output a prompt candidate from the plurality of prompt candidates based on the determined quality of the prompt candidates.
Need to check novelty before this filing date? Find Prior Art